Face Generation¶
In this project, you'll use generative adversarial networks to generate new images of faces.
Get the Data¶
You'll be using two datasets in this project:
- MNIST
- CelebA
Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.
If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".
In [1]:
data_dir = './data'
# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)
In [2]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
Out[2]:
CelebA¶
The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.
In [3]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))
Out[3]:
Preprocess the Data¶
Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.
The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).
Build the Neural Network¶
You'll build the components necessary to build a GANs by implementing the following functions below:
model_inputsdiscriminatorgeneratormodel_lossmodel_opttrain
Check the Version of TensorFlow and Access to GPU¶
This will check to make sure you have the correct version of TensorFlow and access to a GPU
In [7]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf
# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer. You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))
# Check for a GPU
if not tf.test.gpu_device_name():
warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
Input¶
Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:
- Real input images placeholder with rank 4 using
image_width,image_height, andimage_channels. - Z input placeholder with rank 2 using
z_dim. - Learning rate placeholder with rank 0.
Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)
In [8]:
import problem_unittests as tests
def model_inputs(image_width, image_height, image_channels, z_dim):
"""
Create the model inputs
:param image_width: The input image width
:param image_height: The input image height
:param image_channels: The number of image channels
:param z_dim: The dimension of Z
:return: Tuple of (tensor of real input images, tensor of z data, learning rate)
"""
# The real inputs are a tensor of rank 4:
real_inputs = tf.placeholder(tf.float32,
shape=(None, image_width, image_height, image_channels),
name='real_input')
#real_inputs.mark_used()
# The z data is a tensor of rank 2:
z_data = tf.placeholder(tf.float32, (None, z_dim), name='z_data')
#z_data.mark_used()
# alpha, the learning rate, is a real-valued scalar:
alpha = tf.placeholder(tf.float32, name='learning_rate')
return (real_inputs, z_data, alpha)
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)
Discriminator¶
Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).
In [9]:
def discriminator(images, reuse=False):
"""
Create the discriminator network
:param images: Tensor of input image(s)
:param reuse: Boolean if the weights should be reused
:return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
"""
# The following network is based on the network used in the paper
# "Unsupervised Representation Learning With Deep Convolutional
# Generative Adversarial Networks" by Radford, Metz, and Chintala
# It is available on the arxiv server: https://arxiv.org/pdf/1511.06434.pdf
#
# This Deep Convolution GAN (DCGAN) network consists of 5 layers:
# 1. A 5x5 2D convolution layer (with sride of 2) followed by leaky Relu
# 2. Another 5x5 2D convolution with stride of 2, followed by batch-normalization and lRelu
# 3. A 3rd 5x5 2D convolution, this time with a stride of 1, followed by batch-norm and lRelu
# 4. A 4th 5x5 2D convolution, stride of 1, followed by batch-norm and leaky-Relu
# 5. A flattening layer followed by a sigmoid activation function to form the final outputs
# All leaky Relus use the same "leakage" parameter -- essentially the slope of
# the negative half of the Relu
relu_leakage = 0.20 # (as recommended by the DCGAN paper)
keep_prob = 0.6 # For dropout after each lRelu layer
with tf.variable_scope('discriminator', reuse=reuse):
# using 4 layer network as in DCGAN Paper
# Convolution Layer 1
# Reduce image size from 28x28x1 (for MNIST) or 28x28x3 (for celebs - with 3 color channels)
# To 14x14x64
conv_layer_1 = tf.layers.conv2d(inputs = images,
filters = 64,
kernel_size = 5,
strides = 2,
padding = 'SAME',
kernel_initializer = tf.contrib.layers.xavier_initializer())
leaky_relu_1 = tf.maximum(relu_leakage * conv_layer_1, conv_layer_1)
leaky_relu_1 = tf.nn.dropout(leaky_relu_1, keep_prob)
# Convolution Layer 2
# Reduce image size again, to 7x7x128
conv_layer_2 = tf.layers.conv2d(inputs = leaky_relu_1,
filters = 128,
kernel_size = 5,
strides = 2,
padding = 'SAME',
kernel_initializer = tf.contrib.layers.xavier_initializer())
batch_norm_2 = tf.layers.batch_normalization(conv_layer_2, training=True)
leaky_relu_2 = tf.maximum(relu_leakage * batch_norm_2, batch_norm_2)
leaky_relu_2 = tf.nn.dropout(leaky_relu_2, keep_prob)
# Convolution Layer 3
# And, again, to 4x4x256
conv_layer_3 = tf.layers.conv2d(inputs = leaky_relu_2,
filters = 256,
kernel_size = 5,
strides = 1,
padding = 'SAME',
kernel_initializer = tf.contrib.layers.xavier_initializer())
batch_norm_3 = tf.layers.batch_normalization(conv_layer_3, training=True)
leaky_relu_3 = tf.maximum(relu_leakage * batch_norm_3, batch_norm_3)
leaky_relu_3 = tf.nn.dropout(leaky_relu_3, keep_prob)
# Convolution Layer 4
# One last time, to
conv_layer_4 = tf.layers.conv2d(inputs = leaky_relu_3,
filters = 512,
kernel_size = 5,
strides = 1,
padding = 'SAME',
kernel_initializer = tf.contrib.layers.xavier_initializer())
batch_norm_4 = tf.layers.batch_normalization(conv_layer_4, training=True)
leaky_relu_4 = tf.maximum(relu_leakage * batch_norm_4, batch_norm_4)
leaky_relu_4 = tf.nn.dropout(leaky_relu_4, keep_prob)
# Layer 5: Flatten, Logits, and Output
flat_layer = tf.reshape(leaky_relu_4, (-1, 7*7*512))
logits = tf.layers.dense(flat_layer, 1)
outputs = tf.sigmoid(logits)
return (outputs, logits)
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)
Generator¶
Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.
In [10]:
def generator(z, out_channel_dim, is_train=True):
"""
Create the generator network
:param z: Input z
:param out_channel_dim: The number of channels in the output image
:param is_train: Boolean if generator is being used for training
:return: The tensor output of the generator
"""
# The GAN generator generally implements the reverse of the convolution layers:
# De-convolving the outputs to go from 512-to-256-to-128-to-64 pixels
# The final output layer uses a tanh activation function, rather than the sigmoid
# that was used in the discriminator
# As before, use the same relu-leakage slope for all leaky Relus
relu_leakage = 0.20 # As recommended in the DCGAN paper
keep_prob = 0.6 # for dropout
#print ("z-shape=", z.shape)
with tf.variable_scope('generator',
reuse=False if is_train==True else True):
# The generator starts with a fully-connected layer
fc_layer_1 = tf.layers.dense(z, 7*7*512)
fc_layer_1 = tf.reshape(fc_layer_1, (-1, 7, 7, 512))
fc_layer_1 = tf.maximum(relu_leakage*fc_layer_1, fc_layer_1)
####################################
# Perform 3 layers of de-convolution
####################################
# Layer 2
de_conv_layer_2 = tf.layers.conv2d_transpose(inputs = fc_layer_1,
filters = 256,
kernel_size = 5,
strides = 1,
padding = 'SAME',
kernel_initializer = tf.contrib.layers.xavier_initializer())
batch_norm_2 = tf.layers.batch_normalization(de_conv_layer_2, training=is_train)
leaky_relu_2 = tf.maximum(relu_leakage * batch_norm_2, batch_norm_2)
leaky_relu_2 = tf.nn.dropout(leaky_relu_2, keep_prob)
# Layer 3
de_conv_layer_3 = tf.layers.conv2d_transpose(inputs = leaky_relu_2,
filters = 128,
kernel_size = 5,
strides = 2,
padding = 'SAME',
kernel_initializer = tf.contrib.layers.xavier_initializer())
batch_norm_3 = tf.layers.batch_normalization(de_conv_layer_3, training=is_train)
leaky_relu_3 = tf.maximum(relu_leakage * batch_norm_3, batch_norm_3)
leaky_relu_3 = tf.nn.dropout(leaky_relu_3, keep_prob)
# Layer 4
de_conv_layer_4 = tf.layers.conv2d_transpose(inputs = leaky_relu_3,
filters = 64,
kernel_size = 5,
strides = 2,
padding = 'SAME',
kernel_initializer = tf.contrib.layers.xavier_initializer())
batch_norm_4 = tf.layers.batch_normalization(de_conv_layer_4, training=is_train)
leaky_relu_4 = tf.maximum(relu_leakage * batch_norm_4, batch_norm_4)
leaky_relu_4 = tf.nn.dropout(leaky_relu_4, keep_prob)
# Logits and output layer
logits = tf.layers.conv2d_transpose(inputs = leaky_relu_4,
filters = out_channel_dim,
kernel_size = 5,
strides = 2,
padding = 'SAME')
# Reshape final outputs to 28x28 to match original input size
logits_reshaped = tf.image.resize_images(logits, size=(28, 28))
out = tf.tanh(logits_reshaped)
return out
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)
Loss¶
Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:
discriminator(images, reuse=False)generator(z, out_channel_dim, is_train=True)
In [11]:
def model_loss(input_real, input_z, out_channel_dim):
"""
Get the loss for the discriminator and generator
:param input_real: Images from the real dataset
:param input_z: Z input
:param out_channel_dim: The number of channels in the output image
:return: A tuple of (discriminator loss, generator loss)
"""
# Calculate a loss metric for the generator and discriminator set of models
generator_model = generator(input_z, out_channel_dim)
# To get both real and generated discriminator outputs, apply the discriminator
# to both the real input data and the data generated by the generator
disc_real, disc_logits_real = discriminator(input_real)
disc_gen, disc_logits_gen = discriminator(generator_model, reuse=True)
# Use cross-entropy metric to calculate loss of the real and generated images
smoothed_real_labels = 0.95
disc_loss_real = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits = disc_logits_real,
labels = tf.ones_like(disc_real) * smoothed_real_labels))
disc_loss_gen = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits = disc_logits_gen,
labels = tf.zeros_like(disc_gen)))
# The total discriminator loss consists of both the real and generated loss metrics
discriminator_loss = disc_loss_real + disc_loss_gen
# The generator loss metric is just the cross-entropy metric for the generator
generator_loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits = disc_logits_gen,
labels = tf.ones_like(disc_gen)))
return (discriminator_loss, generator_loss)
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)
Optimization¶
Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).
In [12]:
def model_opt(d_loss, g_loss, learning_rate, beta1):
"""
Get optimization operations
:param d_loss: Discriminator loss Tensor
:param g_loss: Generator loss Tensor
:param learning_rate: Learning Rate Placeholder
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:return: A tuple of (discriminator training operation, generator training operation)
"""
# Use the Adam optimizer to optimize the model learning
training_vars = tf.trainable_variables()
discriminator_vars = [var for var in training_vars if var.name.startswith('discriminator')]
generator_vars = [var for var in training_vars if var.name.startswith('generator')]
# Optimize
discriminator_optimizer = tf.train.AdamOptimizer(learning_rate,
beta1=beta1).minimize(d_loss,
var_list=discriminator_vars)
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS,
scope = 'generator')):
generator_optimizer = tf.train.AdamOptimizer(learning_rate,
beta1=beta1).minimize(g_loss,
var_list=generator_vars)
return (discriminator_optimizer, generator_optimizer)
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)
In [16]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np
def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
"""
Show example output for the generator
:param sess: TensorFlow session
:param n_images: Number of Images to display
:param input_z: Input Z Tensor
:param out_channel_dim: The number of channels in the output image
:param image_mode: The mode to use for images ("RGB" or "L")
"""
cmap = None if image_mode == 'RGB' else 'gray'
z_dim = input_z.get_shape().as_list()[-1]
example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])
samples = sess.run(
generator(input_z, out_channel_dim, False),
feed_dict={input_z: example_z})
images_grid = helper.images_square_grid(samples, image_mode)
pyplot.imshow(images_grid, cmap=cmap)
pyplot.show()
Train¶
Implement train to build and train the GANs. Use the following functions you implemented:
model_inputs(image_width, image_height, image_channels, z_dim)model_loss(input_real, input_z, out_channel_dim)model_opt(d_loss, g_loss, learning_rate, beta1)
Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.
In [17]:
def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
"""
Train the GAN
:param epoch_count: Number of epochs
:param batch_size: Batch Size
:param z_dim: Z dimension
:param learning_rate: Learning Rate
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:param get_batches: Function to get batches
:param data_shape: Shape of the data
:param data_image_mode: The image mode to use for images ("RGB" or "L")
"""
##############################################################
# Initialize both the discriminator and generator models
##############################################################
tf.reset_default_graph()
# Use model_inputs to initialize the inputs
input_real, input_z, _ = model_inputs(data_shape[1], data_shape[2], data_shape[3], z_dim)
# Calculate the discriminator and generator loss metrics
discriminator_loss, generator_loss = model_loss(input_real, input_z, data_shape[3])
# Initialize the Adam optimzer for both the discriminator and generator
discriminator, generator = model_opt(discriminator_loss,
generator_loss,
learning_rate,
beta1)
# Keep track of the number of iterations
iter = 0
#############################################################
# Train the model
#############################################################
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for epoch_i in range(epoch_count):
for batch_images in get_batches(batch_size):
# TODO: Train Model
# Normalize batch images to be from -1 to +1
batch_images = batch_images * 2
iter += 1
# Sample z values from -1 to +1
batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim))
# Run the discriminator flow graph
sess.run(discriminator,
feed_dict={input_real : batch_images,
input_z : batch_z})
# And, the generator
sess.run(generator, feed_dict={input_z: batch_z})
if iter % 100 == 0:
# At the end of every 100 iterations, print out the loss metrics
# for both the discriminator and generator
d_loss = discriminator_loss.eval({input_z : batch_z,
input_real : batch_images})
g_loss = generator_loss.eval({input_z: batch_z})
print("Epoch {}/{} \n".format(epoch_i+1, epochs),
"Iteration {} \n".format(iter),
"Discriminator Loss = {:.4f} \n".format(d_loss),
"Generator Loss = {:.4f} \n".format(g_loss))
# Show 25 of the images to see how things are doing
show_generator_output(sess, 25, input_z, data_shape[3], data_image_mode)
MNIST¶
Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.
In [18]:
batch_size = 64
z_dim = 100 # Based on DCGAN paper
learning_rate = 0.0002 # The DCGAN paper recommended 0.0002...
beta1 = 0.50 # The DCGAN paper recommended 0.5 for stability
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 4
mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
mnist_dataset.shape, mnist_dataset.image_mode)
CelebA¶
Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.
In [ ]:
batch_size = 32
z_dim = 100 # Matches feature sizes from DCGAN paper
learning_rate = 0.0002 # Recommended value from the DCGAN paper
beta1 = 0.5 # DCGAN paper indicated this would improve stability
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 1
celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
celeba_dataset.shape, celeba_dataset.image_mode)
Submitting This Project¶
When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.
In [ ]: